On Mon, Mar 6, 2017 at 3:57 PM, Andres Freund <andres@anarazel.de> wrote:
> I'm ok with not immediately doing so, but I think Peter's design isn't
> in line with achieving something like this.
I would be okay with doing this if we had a grab-bag of expensive
checks, that all pretty much require some combination of
ExclusiveLocks and/or ShareLocks anyway, and are amenable to being
combined. I could see that happening, but we're a long way off from
that. When it does happen, we might have other things that are a bit
like bt_index_parent_check(), in terms of locking requirements and
degree of thoroughness, that might themselves require other parameters
specific to the verification that they perform that we cannot
anticipate. For example, maybe bt_index_parent_check() could have a
parameter that is a probability that any given IndexTuple will be
verified against its heap tuple. Then you could have something like a
PRNG seed argument, so you can check different tuples every day. There
are many possibilities, and I hope that eventually amcheck has this
kind of very rich functionality.
When you use the interface you describe to "stack" several checks at
once, less important parameters, like the ones I suggest are going to
be an awkward fit, and so will probably be excluded. I'm not opposed
to having what amounts to a second interface to what bt_index_check()
does, to make this kind of thing work where it makes sense. Really,
bt_index_parent_check() is almost an alternative interface to the same
functionality today. There can be another one in the future, if it
serves a purpose, and the locking requirements are roughly the same
for all the checks. I'd be fine with that. Let's just get the basic
feature in for now, though.
--
Peter Geoghegan
